Real-time Control Design with Resource Constraints
Yifan Wu · 2009
Nowadays, most control systems are manipulated by control algorithms that are implemented as software components. In many practical applications, such software components execute on platforms characterized with constrained resources, such as in embedded systems. To ensure the correct timing behavior of the control software, real-time kernel is employed to schedule several such components running on the same platform and sharing the same limited resources. However, the integration of real-time systems and control systems needs careful analysis and deliberate design with new techniques. In this dissertation, the design problem for real-time control systems in resourceconstrained platforms is investigated. The background knowledge of real-time systems and control systems is introduced, and the state of the art is reviewed. Several new approaches are proposed to contribute to the existing technologies. The limited-preemption scheduling is utilized to improve the responsiveness of control tasks and hence improve the control performance, without jeopardizing the schedulability of the whole system. A general framework for real-time control design is presented where the delay and jitter effect are incorporated into the performance optimization. The resource constraints are characterized using the convex approximation of the EDF deadline space, and a two-step procedure for period and deadline selection is proposed. The utilization of multiprocessor platform is investigated. A search algorithm is presented to exploit the internal parallelism of an application and partition it into several flows that are then implemented using resource reservation to guarantee the timely behavior and provide temporal isolation. The benefits of using this method is shown by a control application involved with a ball-and-plate system.